Neural network trained to control anesthetic doses, keep patients under during surgery

Researchers demonstrate how deep learning could eventually replace traditional anesthetic practices.
Written by Charlie Osborne, Contributing Writer

Academics from the Massachusetts Institute of Technology (MIT) and Massachusetts General Hospital have demonstrated how neural networks can be trained to administer anesthetic during surgery.

Over the past decade, machine learning (ML), artificial intelligence (AI), and deep learning algorithms have been developed and applied to a range of sectors and applications, including in the medical field. 

In healthcare, the potential of neural networks and deep learning has been demonstrated in the automatic analysis of large medical datasets to detect patterns and trends; improved diagnosis procedures, tumor detection based on radiology images, and more recently, an exploration into robotic surgery. 

See also: MIT SCRAM: a new analysis platform for prioritizing enterprise security investments

Now, neural networking may have new, previously-unexplored applications in the surgical and drug administration areas.

A team made up of MIT and Mass General scientists, as reported by Tech Xplore, have developed and trained a neural network to administrator Propofol, a drug commonly used as general anesthesia when patients are undergoing medical procedures. 

In a study due to be published after the virtual-only 2020 International Conference on Artificial Intelligence in Medicine, the team described how they trained algorithms to correctly apply anesthetic doses.

Datasets including patient data which may change recommended levels of anesthetic -- such as weight, age, and preexisting medical conditions -- as well as models that monitor levels of consciousness during a procedure and subsequent recommended drug doses -- were fed into the network.

CNET: Alexa is more vital than ever during coronavirus, and Amazon knows it

As a new exploration of deep learning applications in the medical field, the scientists stuck to a simulated environment and virtual patients. A series of trials were launched to improve the model over time in what is known as a "cross-entropy" method. 

As changes in simulated consciousness were recorded, the model eventually learned how to adapt to neural variations and apply the right dose of Propofol to keep the patient under. 

The neural network has potentially exceeded the teams' expectations. The system now outperforms proportional-integral-derivative (PID) controllers, industry-standard technology used to determine the right levels of and administer anesthetics including Propofol. 

TechRepublic: 2025 forecast: Global IoT healthcare market looks good-a $188.2 billion opportunity

"Deep neural networks allow us to make a model with many continuous input data, so our method generated more coherent control policies than prior table-based policies," Gabriel Schamberg, one of the researchers involved in the study, told Tech Xplore. 

The model is yet to be tried out on live patients and would need to undergo approval for medical trials to take place in controlled clinical settings. If considered a replacement for today's PID controllers in the future, however, neural networking could have the potential to refine what we currently consider the ideal dose of anesthesia for different patients.

Previous and related coverage

Have a tip? Get in touch securely via WhatsApp | Signal at +447713 025 499, or over at Keybase: charlie0

Editorial standards